4  Scientific & Code Review — Seismic Bedload Transport from the Dec-2025 Mt. Rainier Floods

Revisit of the December 2025 “river rumble” work — June 2026

This document revisits the scientific premise, audits the physical equations and their implementation in notebooks/utils.py and scripts/run_river_rumble_batch.py, recomputes the key time-series diagnostics, and assesses whether there is a defensible new finding.

4.1 0. Verdict (TL;DR)

  • The premise is sound and under-exploited. Discharge is measured; the interesting, unclaimed science is in the deviation of the seismic–discharge scaling from the turbulent-flow baseline, and in hysteresis. Nobody has tied the Dec-2025 Rainier floods + debris flows to a mountain-to-sea bedload signal — the niche is real.
  • The “5/4 exponent” target is not correct as stated. It is not a Gimbert/Tsai/Lamb result. The turbulent-flow (water) seismic power scales as P ∝ H^(7/3) ∝ u*^(14/3), which through hydraulic geometry gives P ∝ Q^b with b ≈ 0.9–1.4 (≈ linear). Use the locally-calibrated water exponent (~1) as the baseline; a steeper slope is the bedload signature, not a deviation from 5/4. (5/4 is only a coincidental midpoint.)
  • The recomputed data already show the bedload signature at the Rainier glacial-river stations: at CC.PR03 (Puyallup nr Electron) the slope steepens from b≈1.6 (2–8 Hz) to b≈2.0 (10–30 Hz) — well above the ~1 turbulence baseline, with high correlation. That super-linear, frequency-increasing exponent is a new, defensible result — once the pipeline bugs below are fixed, because the current products are contaminated.
  • The current implementation is a purely empirical band-power-vs-discharge regression. There is no Gimbert/Tsai forward model (no Green’s function, geometric spreading, attenuation, shear-stress or grain-size physics). That is fine as a first pass but the science-plan.md conceptual integral oversells what is coded.

4.2 1. Scientific premise (revisited)

The setting is excellent for the claim. The CC stations sit on Mt. Rainier’s glacial, sediment-charged rivers:

Seismic River / USGS gage Notes
CC.PR01–PR05, CARB, CRBN, GNOB… Puyallup nr Electron 12092000 glacial, debris-flow prone
CC.* Carbon nr Fairfax 12094000 drains Carbon Glacier, very high sediment
CC.* Nisqually nr National 12082500 drains Nisqually Glacier
UW.BHW Snohomish nr Monroe 12150800 large lowland river — negative control

The Dec-2025 atmospheric-river floods and debris flows mobilized large sediment volumes. Because discharge Q(t) is measured, the seismic data are most valuable not for predicting Q but for the residual / scaling-deviation that encodes bedload transport efficiency and supply — exactly as science-plan.md argues. That framing is correct and is the right thing to pitch.

4.3 2. The physics, and the exponent question

4.3.1 2.1 Turbulent flow (water) — Gimbert, Tsai & Lamb (2014, JGR-ES, Eq. 43)

Kolmogorov inertial-subrange turbulence → fluctuating bed tractions → Rayleigh waves with 1/√r spreading and exponential intrinsic attenuation e^(−2πf r /(v_u Q)). The final PSD scales as

P_water(f) ∝ H^(7/3) · sin(θ)^(7/3) · f^(4/3+5ξ) · (path/attenuation term) , with ξ≈0.48.

Key scalings the code should respect: - Depth: P ∝ H^(7/3) (verbatim in GTL14, p. 2221). - Slope: P ∝ S^(7/3) (≈ const at a station). - Shear velocity: u*² = gH sinθ ⇒ P ∝ u*^(14/3). - Distance / Q: enter exponentially, not as a power law. - The model is parameterized in u* and H, NOT in U — a “F ∝ U³” water law is a different simplified convention, not GTL14.

Discharge exponent (the derivation GTL14 omits): with S≈const, P ∝ H^(7/3); at-a-station hydraulic geometry H ∝ Q^f (Leopold–Maddock f≈0.3–0.45; Tsai-2012 supplement gives up to f≈0.6 at fixed width/slope). Hence

P_water ∝ Q^(7/3 · f) ≈ Q^0.9 … Q^1.4, centered near b ≈ 1 (linear).

→ The right baseline is b≈1 (or, better, the measured low-band exponent for that station), not 5/4. GTL14 contains no 5/4, no 14/3, no analytic b at all; they validate numerically per transect against the 2013 Colorado controlled flood.

4.3.2 2.2 Bedload — Tsai, Minchew, Lamb & Ampuero (2012, GRL, Eq. 7, 13)

Grain impacts as impulses → PSD per grain:

P_bed(f,D) ∝ (impact rate) · f³ · m² w² / (ρ_s² v_c³ v_u²) · χ(attenuation) , integrated over the GSD.

Key results: - Linear in bedload flux: P_bed ∝ q_b … - …but ∝ D³ (or D² on rough beds): the signal is dominated by the coarse tail (D₉₄), so a shifting grain-size distribution biases any flux inversion. This is the model’s biggest vulnerability. - Flux itself is threshold-controlled and strongly nonlinear: q_b ∝ (τ*−τ*_c)^1.5 (Fernández-Luque & van Beek). So P_bed vs Q is super-linear and steepens above the entrainment threshold — and is multivalued (hysteretic) in Q. - Caveat (Luong et al. 2024, JGR): the bedrock-saltation model under-predicts flux by 1–2 orders at shallow flow (ignores rolling/sliding).

4.3.3 2.3 Separating the two

  • Frequency: turbulence ~1–10/20 Hz; bedload >15 Hz (analysis bands 30–80 Hz in recent work). The bands overlap — clean separation by frequency alone is approximate.
  • Slope vs Q: turbulence ≈ linear & single-valued; bedload steeper & hysteretic.
  • Hysteresis (Hsu 2011; Roth 2014/2016; Cook 2018 Science GLOF):
    • Clockwise (more power on the rising limb): supply exhaustion, armor break-up, proximal/readily-available sediment.
    • Counter-clockwise (more power on the falling limb): delayed / distal delivery, migrating sediment pulse, bank collapse, tributary input — i.e., precisely the “delayed storage / sediment-pulse” idea worth pitching.
    • Caution (Roth 2016): boundary-roughness change can mimic supply-driven hysteresis.

4.4 3. Implementation critique (prioritized)

4.4.1 Tier 1 — these can flip the conclusions

  1. Silent response-removal fallback → unit mixing. _remove_instrument_response does except: return tr (raw counts). When metadata attach fails on a cached day, band power jumps by orders of magnitude. This is the cause of the absurd dynamic ranges in the saved CSVs (PR03 2–8 Hz spans 7×10²⁶). Fix: fail loudly / drop the day, never silently return counts; verify response attached before integrating.
  2. Lag selection is spurious. estimate_constant_lag_seconds maximizes correlation of log10 levels over ±24 h. On trended storm hydrographs the correlation is dominated by the multi-day trend, so it locks onto large, unphysical lags — the saved fit_parameters.csv has τ = −18 h (proxy leading discharge; the notebook itself flags negative lags as metadata errors). Fix: detrend / first- difference (or band-pass to event timescales) before the lag scan, and restrict to physically plausible lags from gage↔︎station geometry and celerity (~1–4 m/s), with the correct sign for upstream vs downstream gages.
  3. Gage pairing by haversine only. load_station_gage_pairs picks the nearest gage by great-circle distance, ignoring whether it is on the same river or up/downstream. PR01 alone has 9 candidate gages. This injects timing and scaling error. Fix: constrain to the same NHD reach / flowline, record upstream/downstream.

4.4.2 Tier 2 — bias the slope

  1. Whole-day clipping is too aggressive. clip_trace_days_on_stalta_impulses clips an entire UTC day to ±σ if any STA/LTA trigger fires that day. One teleseism nukes a day of real river signal, flattening the slope (plausibly why UW.BHW flow b=0.54 and 10–30 Hz r=0.06). Fix: clip only the triggered windows, not the day.
  2. Regression dilution. OLS of log P on log Q attenuates the slope toward 0 when log Q is noisy/measured-with-error. Fix: report Theil–Sen / Deming (total least squares) and bin-averaged slopes; expect the true exponent to be higher than OLS.
  3. No transport threshold. A single power law over all Q mixes below-threshold (no transport) with above-threshold data. Fit piecewise about an entrainment threshold (the bedrock-diagnostics cell uses an arbitrary 60th-percentile of Q — tie it to τ*_c instead).
  4. Low band under-resolved. Welch nperseg = min(win, 10·sr) → 10 s segments give poor resolution and few bins in the 0.5–2 Hz “flow” band. Use longer segments for low bands.

4.4.3 Tier 3 — robustness / reproducibility

  1. RSS dropna(how="any") drops all times when any component is missing → gaps/bias.
  2. Hysteresis is plotted but never quantified — no index, no per-event separation, so the “delayed storage vs boulder size” hypotheses cannot be tested as written.
  3. Stale, inconsistent products: fit_parameters.csv has only UW.BHW at the spurious −18 h lag, while the figures/CSVs come from different older runs with different cleaning.
  4. No distance/attenuation correction ⇒ absolute power is not comparable across stations; only per-station relative change is interpretable. (science-plan.md admits this.)
  5. No physical forward model — the proxy is band-integrated PSD in arbitrary units; there is no path to absolute flux without GTL14/Tsai inversion. State this plainly.

4.5 4. Recomputed diagnostics (Dec-2025 event)

Recomputed from the existing aligned CSVs with robust 6·MAD log-space outlier rejection, zero lag (the saved −18 h lag is spurious), Theil–Sen + OLS slope, and a Lawler hysteresis index over ±3 days around the event peak. Saved to notebooks/data/results/RECOMPUTED_diagnostics_2026review.csv.

Station / band r slope b HI (event) Q_peak (m³/s) interpretation
CC.PR03 2–8 Hz (Puyallup) 0.93 1.57 −0.05 323 above turbulence baseline
CC.PR03 10–30 Hz 0.81 2.06 +0.03 323 bedload regime; slope steepens with f
CC.PR02 (single band) 0.75 1.95 +0.07 (CW) 323 super-linear; mild clockwise
UW.BHW 0.5–2 Hz (Snohomish) 0.56 0.54 n/a 3000 shallow — large lowland river / contamination
UW.BHW 10–30 Hz 0.06 0.07 — 3000 no bedload signal (good negative control)

(These are the pre-fix, as-found numbers; the bedload signal shrinks after the Tier-1 fixes — see the clean values in the manuscript.)

Update (2026): UW.BHW (lowland Snohomish) was used here only as an early negative control. It was subsequently removed from the analysis as too far from any bedload source to be informative; the published transect uses the internal source→downstream decay (CC.PR0x → CC.SIFT → CC.TRON) as the contrast instead. This chapter retains UW.BHW as a record of the development process.

Event peak at Puyallup nr Electron: 2025-12-09 ~03:30 UTC, ≈323 m³/s (11,400 cfs).

4.6 5. Is there a new finding?

Yes — a candidate, with caveats. At the Rainier glacial-river stations the high-frequency seismic power scales super-linearly with discharge (b≈2) and the exponent increases with frequency (1.6 → 2.0 from the flow band to the bedload band), far above the ~1 turbulent-flow baseline. That frequency-dependent steepening is the textbook bedload signature (Tsai 2012: P_bed ∝ q_b with q_b ∝ (τ*−τ*_c)^1.5, plus D³ grain leverage). The lowland negative control (UW.BHW, r≈0.06 in the bedload band) confirms the high-frequency response is not a generic flow artifact. Differing hysteresis sign between stations hints at event-scale changes in transport efficiency / a migrating post-debris-flow pulse — the “delayed delivery” story.

The honest caveat: these slopes are computed from products contaminated by the Tier-1 bugs (unit-mixing from silent response failures, spurious −18 h lag, day-clipping). The numbers are suggestive and physically coherent but not yet publication-grade. The finding becomes solid only after re-running with the fixes in §3.

4.8 7. Fluvial Hydrologist review

Committed (424a115) — the Nisqually two-basin result is integrated and the book renders clean. The migration-scales-with-break-steepness relationship (PR01 −50 m/Δb+1.7; UW.LON −240 m/Δb+3.1) is now a real, if two-point, argument that these breaks are geometric.

Now, switching hats. Here is my honest evaluation as a fluvial hydrologist / geomorphologist asked to referee this work.

Does this work generate new fluvial-science knowledge? A skeptical reading Bottom line up front: This is a strong observational-methods contribution. As fluvial-process science, most of what it reports is the recovery of already-known fluvial behavior with a new sensor, not new knowledge about how rivers work. There is one genuinely novel caution for a subfield, one tantalizing-but-unproven candidate for new process insight, and a large untapped opportunity that, if exploited, would be the real fluvial-science contribution.

What is NOT new fluvial knowledge (most of it) P ∝ Q^b turbulent-flow scaling — established (Gimbert 2014; Bakker 2020; Shakti & Sawazaki 2021). New site, not new physics. Virtual discharge / distributed gaging — Shakti & Sawazaki (2021) already did this for ungauged basins. Network density is incremental. Rain-on-snow generated the floods (warm-AR thread) — textbook PNW hydrometeorology. The SNOTEL panel confirms; it doesn’t teach a hydrologist anything new. ~36 h upstream-to-downstream lead — this is essentially the kinematic flood-wave travel time from Electron (RM 41) to Puyallup (RM 10), which any hydraulic-routing model gives. New sensing of a known celerity, not a new hydrological insight. Braided channels migrate / avulse and the active thread controls the local signal — this is the definition of a braided river. A geomorphologist already knows the source is distributed and non-stationary. The one genuinely useful (semi-novel) point — but it’s aimed at seismologists, not us The thesis that a seismic P–Q break is a fluvial regime transition (geometry), not necessarily a bedload-transport onset is a real corrective — but it corrects an over-interpretation made by the environmental-seismology community, not a gap in fluvial knowledge. We already know that at-a-station width/depth partitioning, overbank activation, and anabranch switching produce non-linear stage–hydraulics. So this is valuable methodological hygiene for fluvial seismology, and the “braided reach is a distinct calibration regime” caution is a fair, modestly-new point for that small subfield. It is not a discovery about rivers.

The tantalizing candidate that is not yet knowledge The AR2 “de-armoring / cross-pulse supply” effect (most bedload mobilized in AR2 despite AR2 not being the largest pulse) would be a genuine sediment-supply insight — supply-limited hysteresis across a compound event is real geomorphology. But the paper itself concedes the band is turbulence-dominated and the signal may be geometric, so this stays a bounded hypothesis. As written, it’s a hint, not a result. The migration-vs-break-steepness relationship is the most novel quantitative claim, but it is n = 2, with an uncontrolled seasonal (post-flood January) confound and a threshold-sensitive magnitude. Suggestive, not established. Where the real new fluvial knowledge is hiding (and is currently left on the table) The seismic network’s unique affordance is continuous, sub-daily, spatially-distributed observation — something neither gages nor satellites provide. Braided-reach reorganization is normally seen only as before/after snapshots (your own Sentinel test) or inferred post-hoc. With 10-minute seismic illumination at several stations you could, in principle, answer questions a geomorphologist cannot currently answer:

When during the hydrograph does the active thread switch/avulse — on the rising limb, at peak, or on recession? Does channel reorganization lead or lag peak discharge? Does the avulsion propagate through the cluster (a migrating front) and at what celerity? Is the cross-AR baseline drift monotonic or stepwise (continuous lateral reworking vs discrete avulsion events)? Resolving the timing and kinematics of braided-channel reorganization at event scale would be a genuine geomorphic discovery. The data already contain it (the baseline-drift timing, the per-AR loops); the paper currently reports the existence of migration, not its dynamics.

Verdict As environmental-seismology / remote-sensing methods: solid, publishable, JGR-ES-appropriate — distributed glacial-river monitoring + a seismic–satellite discrimination of geometric vs bed-mechanical breaks. As fluvial-geomorphology knowledge: thin as framed. It confirms known process with a new instrument and issues a useful caution to one subfield. To make it a fluvial-science contribution: pivot the spine from “what the break is” (a static classification) to “the event-scale dynamics of braided-channel reorganization, resolved seismically” — i.e., use the continuous timing the sensor uniquely provides. That reframes the work from “a clever new gauge” to “a new way to watch a braided river rearrange itself during a flood.”

4.9 8. New result — TIMING of braided-channel reorganization (the geomorphic angle, tested)

Workflow workflows/21_braided_reorg_timing.py --basin puyallup|nisqually → config/braided_reorg_timing_puyallup.json / _nisqually.json + paper/figures/fig22_braided_reorg_timing[_nisqually].png. This is the §7 recommendation, executed.

Method. The cross-AR baseline drift (workflow 18’s baseline_offset, previously only 3 per-AR medians) is resolved continuously by tracking the seismic residual r(t) = log10 P − (a + b·log10 Q) at matched discharge: for reference levels the compound hydrograph crosses repeatedly during the event (60–160 m³ s⁻¹ — inter-pulse troughs + limbs), sample r at every crossing and reference each level to its own rising-limb value. The pooled series c(t) is a sequence of same-stage snapshots, so any change is channel/source geometry, not discharge. Onset = first sustained >2σ departure; the dominant step is a logistic fit (midpoint t50, width 4τ, magnitude Δ) with a bootstrap CI. (The matched-LOW-FLOW baseline, Q<median, only brackets the step — Q never returns to baseflow between pulses — so the matched-Q crossings are what resolve the transition.)

Result — reorganization LAGS peak discharge; it never leads. Peak Q at Electron is 2025-12-09 03:30 UTC. The rising-limb and pre-flood baselines are flat. The drift emerges on the falling limb of the main pulse and the later recessions:

Station onset (lag vs peak) dominant step t50 (lag) end-state (log₁₀) character R² verdict
CC.PR01 — — −0.01 — 0.34 reversible / transient — thread migrated and re-occupied (no persistent step)
CC.PR02 12-09 16:25 (+13 h) 12-12 01:31 (+70 h, CI +69…+72) +0.30 stepwise (≈11 h) 0.66 persistent (+)
CC.PR03 12-09 23:50 (+20 h) 12-12 05:06 (+74 h, CI +72…+75) +0.21 stepwise (≈12 h) 0.72 persistent (+)

Three genuinely new, gage-/satellite-invisible statements: 1. Lead/lag: channel reorganization lags peak discharge — onset ~13 h into the recession of the main pulse, the persistent step ~3 days later on the final recession. It is a recession phenomenon (falling-stage bank collapse / thread abandonment), not a rising-limb or at-peak one. 2. Stepwise, not gradual: the persistent shifts complete in ~half-day transitions (τ ≈ 3 h), i.e. discrete avulsion/thread-switch events, not continuous lateral reworking. 3. Spatial heterogeneity over ~1–2 km: PR01 is reversible (the satellite-confirmed big-migration site moved and partly returned) while PR02/PR03 are irreversible — the braided reach does not reorganize as one body. The PR02→PR03 persistent steps are near-synchronous (Δt≈4 h, within timing uncertainty), so the data show a recession-triggered local response, not a resolvable laterally-propagating front.

Honest caveats. n = 3 stations; the compound hydrograph densely samples matched-Q only at ~3 epochs (AR1 limbs, the AR1/AR2 trough, the final recession), so t50 is pulled toward the largest (final-recession) increment — the onset (+13 h) is the more robust “when it starts,” t50 (+70 h) the “when it’s mostly done.” Band is 5–15 Hz (turbulence + low bedload); the residual is geometry-dominated by construction but a within-event roughness change could co-vary. Still: this is the first event-scale timing of braided reorganization from the array, and it is the defensible seed of the reframed fluvial-science spine.

4.9.1 8b. Second basin — Nisqually (UW.LON, CC.GTWY): the clean signature does NOT reproduce

Same diagnostic, paired to Nisqually nr National 12082500 (main peak 2025-12-10 14:15 UTC, 425 m³ s⁻¹ — here the second AR pulse is the larger one, so the rising-limb reference uses the first pulse and lead/lag is measured against the later peak). The bounded, persistent positive Puyallup step (+0.2…0.3 log, on the recession) does not appear:

Station b matched-Q behavior end-state (log₁₀) R² verdict
UW.LON 2.7 crashes ~1.5–1.8 log on the AR1 recession, recovers on the AR2 rise −0.48 0.05 reversible / unresolved — no persistent step (hysteresis/timing on a steep reach)
CC.GTWY 1.6 large monotone negative drift after the peak −1.50 0.51 persistent (−) but ⚠ supply-confounded

Why it differs — the rain→snow supply shutoff (mechanism, per MD). Matched-Q controls for discharge but not for what sources that discharge. After the warm AR the freezing level drops and precipitation falls (and stays) as snow, so drainage-wide runoff — and the sediment it mobilizes — collapses (Q falls 425 → ~40 baseflow). Storm flow is sediment-loaded (loud); the post-event snowmelt-free recession delivers the same Q sediment-starved (quiet). So the bed goes quiet at equal Q for a supply reason, not a geometric one. The workflow measures this basin-wide as the raw post-event decline (supply_shutoff_decline, here ≈ −0.7 log at GTWY, −1.1…−1.6 log at the Puyallup stations) and auto-flags any negative persistent step that is same-signed with it. GTWY’s −1.5 trips the flag → it cannot be cleanly attributed to channel reorganization. The higher, colder Nisqually converts more of its catchment to snow, so the shutoff is the dominant post-event signal there.

Why the Puyallup result survives this and the Nisqually one doesn’t. The supply shutoff is negative; the Puyallup steps are positive — opposite sign — so the geometric step is measured against a headwind and is conservative (the true step is if anything larger). At Nisqually the candidate step is itself negative, the same sign as the shutoff, so the two are inseparable. Net: the event-scale reorganization-timing diagnostic is cleanly interpretable in the low-elevation, sediment-charged braided Puyallup reach, but confounded at the higher, snow-dominated Nisqually — a real methodological boundary that reinforces the PR cluster (satellite-confirmed) as the clean geometric end-member. UW.LON’s large reversible swings are a separate caution: on a steep (b≈2.7) reach the matched-Q residual is sensitive to limb/timing and supply hysteresis, so its steep P–Q break is not the braided-geometric end-member PR01 is.

4.9.2 8c. Lag-corrected cross-check — the Nisqually swings are largely a gage-geometry artifact

workflow 21 now estimates a constant P–Q flood-wave lag per station (limb-timescale cross-correlation, 36 h high-pass to avoid trend-locking) and shifts the gage discharge to the station before forming the residual — but only where a real travel time exists (gage ≥3 km away); at a co-located gage the estimated lag is sub-resolution noise and is not applied. The implied celerities are physical and validate the correction:

Station gage distance lag implied celerity effect on matched-Q
CC.PR01/02/03 0.2–1.9 km (co-located) ~0 (not applied) — none — result unchanged
CC.GTWY 12.8 km downstream −105 min 2.0 m/s end-state −1.50 → −0.57
UW.LON 20.9 km downstream −180 min 1.9 m/s swing span 2.1 → 1.6; still unresolved

The Puyallup gage (Electron) is essentially on the PR cluster, so the clean result needs no correction. The Nisqually gage (National) is 13–21 km downstream, so gage Q lags the discharge passing the station by the wave travel time; correcting it collapses GTWY’s apparent −1.5 drift to −0.57 — now within the rain→snow supply-shutoff envelope (−0.7), i.e. no geometric signal survives — and shrinks UW.LON’s swings, which remain unresolved (sigmoid Δ disagrees with the end-state sign → flagged reversible). A constant lag cannot fully fix a 20 km path whose celerity varies with stage, so the far-gage Nisqually matched-Q is intrinsically limited — the lag check converts “the method fails at Nisqually” into the sharper, correct statement: the Nisqually gage is too far downstream to time a 10-min source process against.

4.10 9. Why the two basins differ — hydrology & geomorphology of the two source reaches

Requested review. Tagged by confidence: ✔ = verified from data this session; ◐ = inferred / needs a quick check. The throughline: the very properties that make the Nisqually a more dramatic braided site also make it harder to read seismically against a distant gage.

Property Puyallup head (CC.PR01/02/03) Nisqually (CC.GTWY, UW.LON) Consequence for the seismic method
Gage geometry ✔ gage co-located (0.2–1.9 km) gage 13–21 km downstream the single biggest difference: clean vs lag-confounded matched-Q (§8c)
Active-channel width ✔ 17–44 m (optical wet width) 190–624 m at UW.LON (×3.3 Nov→post) wide braidplain ≫ compact channel → the line-source / fixed-r model fails worse; source is hundreds of m of shifting anabranches
Wet area in AOI ✔ 52,600 m² 249,000–503,000 m² source spread over ~5–10× the area → far more non-stationary geometry
Station elevation ✔ 461–648 m 617–853 m higher Nisqually converts more of its catchment rain→snow → larger supply-shutoff confound
Peak discharge (Dec-2025) ✔ 323 m³ s⁻¹ (Electron) 425 m³ s⁻¹ (National) bigger flow; combined with the wide section → deeper hydraulic-geometry effects
Scaling exponent b ✔ 1.1–1.9 (PR cluster) 2.7 (UW.LON, steepest), 1.6 (GTWY) steep b → matched-Q residual hypersensitive to small Q/timing error (amplifies the gage-lag artifact)
Predicted geometric drift ✔ +0.07…+0.28 log (PR01 largest) +0.51 log (UW.LON) Nisqually’s geometric change is larger, not smaller — it is unmeasurable cleanly, not absent
Drainage flank / aspect ◐ NW flank of Rainier S flank different storm exposure, freezing-level behavior, glacier melt timing
Source glaciers / network ◐ Puyallup + Tahoma + S. Tahoma Glaciers; several merging threads Nisqually Glacier (+ Wilson); different confluence topology more/fewer distributed inputs change source stationarity
Drainage area ◐ smaller (consistent with lower peak Q) larger (consistent with 425 vs 323 m³ s⁻¹) larger basin → longer, more dispersive flood wave to the gage
Valley confinement ◐ narrower, more incised/confined reach wide, unconfined outwash plain (consistent with the width data) confinement keeps the Puyallup source compact and stationary

Synthesis. The Puyallup PR cluster is a Goldilocks reach for source-resolved river seismology: a compact (tens-of-m) channel with a co-located gage, low enough to stay rain-driven through the event. The Nisqually stations sit on a wide (hundreds-of-m) high- elevation braidplain gauged 13–21 km downstream — every one of those differences (width, elevation/snow, gage distance, steep b) pushes the matched-Q diagnostic from clean toward confounded, even though the physical reorganization there is larger. This is not a failure of the idea; it is its domain of applicability, and stating it is a genuine methods contribution (which braided-reach competitors — Gangemi 2026 — have not made).

4.11 10. Master update plan — figures, threads, and the spine ⟦TAG: REORG-2BASIN-2026-06-25⟧

Everything below is the plan of record for the next pass. Commit-tagged so progress is trackable. Principle the user set: every thread T1–T13 ships — as the main spine or as supplementary — so the question is allocation and figure-readiness, not what to cut.

4.11.1 10a. Figure-by-figure update checklist

Status: ✅ current · ⚠ needs refresh to the NWIS/flood-window/lag numbers · ➕ new this session.

Fig Update needed
fig1 map ✅ DONE — gage→station connectors + distance labels (“gage co-located” at Puyallup; “13/21 km to gage” for GTWY/UW.LON)
fig2 scaling-b ✅ confirmed lag-robust — applying the far-gage lag moves UW.LON/GTWY b by <0.05 (within CI), so the break markers + b values stand uncorrected
fig3 P–Q scatter ✅ DONE — far-gage (UW.LON/GTWY) Q lag-aligned before the fit; panel title tags the shift; slope essentially unchanged
fig4 hysteresis ✅ DONE — Nisqually loops re-drawn lag-corrected (this is where the 13–21 km lag mattered most)
fig5 event timeseries ✅ keep; annotate AR1/AR2/AR3 + warm/cold tail
fig6/7/8 bedload(t), per-AR, b(t) ⚠ re-confirm against complete NWIS; label band as turbulence+low-bedload
fig9 attenuation ✅ supporting
fig10 early-warning ⚠ recompute lead with lag-aware celerity (we now have 1.9–2.0 m/s in-basin)
fig11 spectra / fig13 30–50 Hz ✅ the frequency-limit (T4) figures
fig12 virtual-Q ⚠ note braided-drift + lag caveat on the Nisqually virtual gage
fig14 threshold Qc ✅ two-exponent + ΔBIC
fig15 rating ✅
fig16 braided hysteresis ⚠ supersede/extend with the new timing view (fig22)
fig19 braid satellite ✅ DONE — both regions cached + offline-rebuildable; the Nisqually-beside-Puyallup view is now its own figure (fig24)
fig21 warm-AR snow ⚠ retitle to the rain→snow supply-shutoff mechanism; tie to fig22 confound
➕ fig22 reorg-timing ✅ DONE (×2 basins) + flood-wave lag correction — lead the geomorphic-dynamics result
➕ fig23 domain panel ✅ DONE — two-basin domain-of-applicability (gage-distance × channel-width × elevation → clean/confounded); workflows/22_domain_panel.py
➕ fig24 two-region braid ✅ DONE — Puyallup (compact incised) vs Nisqually (wide braidplain) change maps side-by-side, offline from cache; workflows/23_braid_two_region.py
figS traffic / GIFs ✅ supplement

Spine LOCKED (2026-06-25): the dynamics-led Spine 1 is now the chosen spine (see paper/ROADMAP.md → “✅ SPINE LOCKED”). Remaining ⚠ items (fig5/6/7/8/10/12/16/21 refreshes) are cosmetic/labelling against already-correct numbers; the load-bearing figures (fig1, fig3, fig4, fig22, fig23, fig24) are done.

Reproducibility hardening (done this pass). figures-from-cache now rebuilds every figure offline, including the satellite fig19/fig20: workflow 19 caches its derived Sentinel mndwi/active-channel rasters to committed notebooks/data/braid_cache/*.npz (+ station pixels json, December-series csv) and replays them with --from-cache, so the flaky Planetary Computer API is never needed to rebuild the paper. Workflows 21 (both basins) + 22 are wired into the offline Makefile path. Only raw seismic (slow) and a live PC refresh remain on the repro path.

4.11.2 10b. Thread disposition (all 13 ship)

  • Spine-eligible (lead candidates): T9 braided source-model breakdown · NEW reorg-timing (event-scale dynamics) · T5 two-regime break · T4 frequency rigor.
  • Framing / payoff (main text, supporting): T1 scaling backbone · T2 two-basin (now the domain contrast) · T7 virtual-Q field · T8 ~36 h early warning · T10 the event · T13 warm-AR rain→snow (now also the supply-shutoff confound).
  • Methods / supplement: T6 attenuation · T11 detectability map · T12 future high-rate · T3 time-dependent bedload (bounded hypothesis, supplement).

4.11.3 10c. Three candidate spines (pick one; the other two become section arcs)

SPINE 1 — “Watching a braided river rearrange itself during a flood” (geomorphic-dynamics-led). ★ recommended. Lead with the event-scale reorganization TIMING (fig22): seismics resolve when the active thread reorganizes — it lags peak, is recession-driven and stepwise — which gages and before/after satellites cannot. T9 (source-model breakdown) + T5 (the static break signature) + the two-basin domain result (§9, why it is clean at Puyallup, confounded at Nisqually) form the body; everything else supplements. Most novel; matches the referee’s “pivot to dynamics” and the open gap (Gangemi 2026 declines a braided model). Journal: JGR-ES.

SPINE 2 — “The domain of source-resolved river seismology on a glacial braidplain” (rigor/methods-led). Headline = honest boundaries: T4 (frequency/Nyquist) + T9 (braided source) + the method-domain law (gage co-location, channel width, snow fraction set where matched-Q / virtual-Q works) + T11 (detectability). Turns the negative Nisqually result and the lag/width/snow analysis into the contribution. Most defensible; least “exciting.”

SPINE 3 — “A seismic network as a distributed flood & sediment observatory” (application/hazard-led). T1+T2+T7 virtual-Q field + T8 early warning for the lahar-prone corridor, with reorg-timing + braided + the domain result as the reliability caveat. Most fundable; the survey flagged the virtual-gage concept as the most saturated (Shakti & Sawazaki 2021), so it must be framed as density+dynamics, not concept.

4.11.4 10d. Execution order (next session)

  1. Regenerate all ⚠ figures from the complete-NWIS + flood-window + lag pipeline; verify numbers.
  2. Build the proposed fig23 domain panel (the §9 table as a figure).
  3. Lock Spine 1; demote T1/T2/T7/T8 to framing/payoff; move T3/T6/T11/T12 to supplement.
  4. Reconcile manuscript text (Abstract/Intro/Discussion/Conclusions) to the dynamics spine.
  5. Out-of-sample March-2026 event as the robustness close (break recurrence + lead + rating transfer).